Thermal etching of AlF3 and thermal atomic layeretching of Al2O3

Cite as: J. Vac. Sci. Technol. A 38, 022603 (2020); doi: 10.1116/1.5135911

View Online Export Citation CrossMarkSubmitted: 8 November 2019 · Accepted: 30 December 2019 ·Published Online: 14 January 2020

Andreas Fischer,1,a) Aaron Routzahn,1 Younghee Lee,1 Thorsten Lill,1 and Steven M. George2

AFFILIATIONS

1Lam Research Corporation, 4400 Cushing Parkway, Fremont, California 945382Department of Chemistry, University of Colorado at Boulder, Boulder, Colorado 80309

Note: This paper is part of the Special Topic Collection Commemorating the Career of John Coburn.a)Electronic mail: andreas.fischer@lamresearch.com

ABSTRACT

Thermal etching of AlF3 with dimethyl-aluminum chloride (DMAC) and thermal isotropic atomic layer etching (ALE) of Al2O3 withalternating anhydrous hydrogen fluoride (HF) and DMAC steps were studied. DMAC vapor etches AlF3 spontaneously at substrate tempera-tures above 180 °C. The thermal etching reaction of AlF3 with DMAC exhibited no self-limitation and showed a linear dependence onDMAC pressure. The authors determined an activation energy of 1.2 eV for this reaction. When Al2O3 is fluorinated, DMAC removes thefluorinated layer partially. The etch amount per cycle (EPC) in thermal isotropic ALE of Al2O3 with HF/DMAC is primarily determined bythe fluorination step placing significant importance on its design. Fluorination with HF gas was found to be more effective and repeatablethan with NF3. Plasma fluorination is faster and provides higher EPC, but the selectivity to Si3N4 or SiO2 mask materials is compromised.For pressures between 10 and 110 mTorr and a substrate temperature of 250 °C, thermal ALE of Al2O3 with HF/DMAC was found to havea very high selectivity to SiO2 and amorphous silicon. HfO2, however, etched with similar EPC as Al2O3.

Published under license by AVS. https://doi.org/10.1116/1.5135911

I. INTRODUCTION

In the mid-1970s, John Coburn began the study of reactiveion etching (RIE) together with his co-worker Harold Winters.The focus of their work was on the surface science aspects of theinvolved mechanisms.1 In 1979, they published their seminal paperdescribing the synergy between ions and neutrals in a beam experi-ment with XeF2 neutrals and argon ions.2 The combination of thetwo beams produced a large enhancement in the material removalrate due to the mechanism of RIE. In a modern-day interpretationof their result, one could say that had they alternated between theXeF2 neutral and argon ion beams, they would have produced thefirst directional atomic layer etching (ALE) process. Winters andCoburn described the etching process as a sequence of five steps:physisorption, dissociative chemisorption, chemical reaction withthe surface, desorption of the reaction product, and removal ofnonreactive residues if they exist.2,3 The last two steps are accom-plished by ion bombardment in RIE and in directional ALE.

In this paper, we report on using thermal energy and a ligandexchange reaction instead of ion bombardment to stimulate desorp-tion. ALE processes that utilize radicals and neutrals in self-limited

modification and removal steps are called thermal isotropic ALE.4

Thermal isotropic ALE has recently garnered much attention bythe ALD and etching communities. It was first reported for etchingof Al2O3 by exposure to alternating hydrogen fluoride (HF) andtin-acetylacetonate [Sn(acac)2] vapor.

5 At least four classes of thermalisotropic ALE are known to date: ligand exchange ALE,5 conversionALE,6 oxidation/fluorination ALE,7 and chelation/condensationALE.8,9 In this work, we studied thermal isotropic ALE using aligand exchange reaction. The Al2O3 surface is fluorinated in a firststep and removed with dimethyl-aluminum chloride (DMAC) in asecond step. This reaction as well as the related Al2O3 ALE withHF/trimethylaluminum (TMA) were reported previously.10

Thermal ALEs of Al2O3 with HF/Sn(acac)2 and HF/TMA areconsidered archetypical systems for thermal isotropic ALE. For bothreactions, the etch amount per cycle (EPC) increases with the tem-perature, but the underlying mechanisms appear to be different. Leeet al. found that for HF/Sn(acac)2, acetylacetonate (acac) containingcompounds such as SnF(acac) remain at the surface.11 The Al2O3

EPC is inversely dependent on the coverage with acetylacetonatesurface species, which was found to be lower at higher temperatures.

ARTICLE avs.scitation.org/journal/jva

J. Vac. Sci. Technol. A 38(2) Mar/Apr 2020; doi: 10.1116/1.5135911 38, 022603-1

Published under license by AVS.

This behavior suggested that acetylacetonate surface species mayhave a site-blocking effect.11 The depth of fluorination can be deeperthan the EPC and the latter is determined by the site-blocking effectof acac containing compounds. These compounds must be removedby HF in the subsequent surface modification step of the ALE cycle.Sn(acac)2 will not etch AlF3 in a continuous process.12

TMA removes AlF3 without site blocking. Cano et al. reportedproportionality between the Al2O3 EPC and the aluminum fluoridelayer thickness for thermal isotropic ALE with HF/TMA.13 Thethickness of the fluorinated layer was measured with XPS andmodeled as Al2OF4 or a combination of AlF3 and AlOF. The fluori-nated layer thickness was found to be roughly two times the EPC,the latter increased with temperature and HF pressure. This mecha-nism for the fluorination of Al2O3 is similar to the mechanismfor silicon oxidation.13,14 The fluorination step is critical for Al2O3

ALE with TMA/HF because it determines the EPC. The TMAremoval step is essentially a thermal etching process of AlF3. Towhat degree TMA etches incompletely fluorinated aluminumAlOxFy is not known.

In this paper, we explore the relationship of thermal etchingof fully fluorinated AlF3 with DMAC as well as thermal ALE ofAl2O3 with NF3/DMAC or HF/DMAC.

II. EXPERIMENT

The experiments were conducted in a modified RIE chamberfor 300 mm wafers as shown in Fig. 1. The wafer was held by an

electrostatic chuck (ESC), which heated the wafer to temperaturesbetween 150 and 280 °C. The heat was transferred via helium gasbetween the ESC and wafer backside surfaces allowing adequatewafer heating even at process pressures in the millitorr range.The chamber is equipped with a turbomolecular pump capable tomaintain pressures between 1 and 500 mTorr. DMAC was con-verted from liquid to gas in a vaporizer and drawn by the processchamber vacuum to the substrate. HF and other gases were deliv-ered from a gas cabinet. All gases were injected through a gasnozzle above the center of the wafer. Flows were between 50 and500 SCCM for DMAC vapor and all process gases. Plasma wasgenerated with a transformer coupled plasma source with an excita-tion frequency of 13MHz,15 resulting in ion energies in the rangebetween 10 and 20 eV. The gap between wafer and the top windowof the reactor was 15 cm. While such a large gap required severalseconds long pump steps to ensure minimal intermixing of HF andDMAC in the gas volume above the wafer (optimal pump step timesof 30 s were determined experimentally by maximizing EPC with allother process parameters kept constant), it allowed us to utilize aplasma source that is fully optimized for ion flux uniformity and ionenergy. The reactor walls were covered with Y2O3; the exact details ofthe coating process and thickness are proprietary to the LamResearch Corporation.

Some early experiments were conducted in a miniaturizedversion of this chamber for processing of 100 mm wafers andcoupons. There, all reactor walls were stainless steel and the topwindow was made of yttria-coated alumina. The samples were placedon a grounded pedestal without ESC.

We used 2 × 2 in. coupons that were gallium-bonded to300 mm bare silicon carrier wafers with either 24 nm thick AlF3 or28 nm thick Al2O3 deposited by ALD on top of SiO2. In addition,patterned Al2O3 samples with an Si3N4 mask featuring 60 nm wideholes were tested.

A Thermo Fisher Scientific Theta 300 standalone XPS toolwith a monochromatic Al kα x-ray source at 1486.7 eV was utilizedto characterize the surface composition. There was a queue time inair of about 30 min between sample processing and XPS measure-ments. Film thicknesses were measured with spectral ellipsometryand SEM cross sections.

III. RESULTS AND DISCUSSION

A. Thermal etching of AlF3

At a pressure of 30 mTorr and a wafer temperature of 250 °C,DMAC etched AlF3 spontaneously as shown in Fig. 2. The gas wasdelivered in time increments of 30 s. The etched amount of AlF3 isa linear function of time indicating that there is no self-limitationof the etch due to an accumulation of reaction products at thesurface or other impeding processes. The original AlF3/siliconinterface was reached after nearly 50 DMAC steps of 30 s each,resulting in etch stop due to the complete removal of AlF3. Wedetermined an etch rate of 1.04 nm/min for AlF3. Figure 3 depictsthe pressure dependence of the etching rate of this material.We found 1.50 and 2.23 nm/min at 60 and 100 mTorr, respectively.A linear trendline fits the data that indicate that the reaction is afirst order surface reaction in the explored pressure range. A linear

FIG. 1. Experimental setup: (1) 300 mm wafer, (2) electrostatic chuck, (3) turbopump, (4) source RF power, (5) gas injection, (6) dielectric window, (7) inductiveantenna, (8) chamber walls, (9) vaporizer, and (10) gas line from the gas box.

ARTICLE avs.scitation.org/journal/jva

J. Vac. Sci. Technol. A 38(2) Mar/Apr 2020; doi: 10.1116/1.5135911 38, 022603-2

Published under license by AVS.

fit also suggests that the DMAC surface coverage does not changesignificantly in this pressure region.

The surface reaction is strongly temperature dependent. Below180 °C, we observed no etching but an increase in film thicknessas measured by spectral ellipsometry. The composition of thislow-temperature film was not analyzed. This film may result fromthe formation of mixed halide complexes between DMAC and AlF3.Above 180 °C, etching occurred with an exponential rate increaseversus temperature (see Fig. 4). An activation energy of 1.2 eV can becalculated from the three measured etching rates above 180 °C usingthe Arrhenius approach. The corresponding Arrhenius plot is shownin Fig. 5.

B. Fluorination of Al2O3

Thermal etching can be used to design thermal isotropic ALEprocesses if two conditions are met. The removal precursor mustetch the modified material but not the original one, and a thermalreaction must exist, which converts the surface from the original tothe modified material in a self-limited regime. DMAC has beenapplied in the removal step of fluorinated Al2O3 previously.10,16

Because DMAC etches AlF3 thermally, the EPC of the reaction isdetermined by the depth of fluorination. The fluorination step istherefore critically important for this reaction.

We explored various fluorination methods using NF3 andanhydrous HF via thermal fluorination and with plasma. Figure 6

FIG. 2. Thermal etching of AlF3 with DMAC: etched amount vs number ofDMAC cycles where each cycle is 30 s long.

FIG. 3. Thermal etching of AlF3 with DMAC: etching rate as a function ofpressure.

FIG. 4. Thermal etching of AlF3 with DMAC: etching rate as a function of wafertemperature. The large error bar at 250 °C likely stems from repeatability issuesin heat transfer between the carrier wafer and the AlF3 coupon.

FIG. 5. Arrhenius plot for thermal etching of AlF3 with DMAC. Vertical axisshows the natural logarithm of the etch rate measured in nanometers perminute.

ARTICLE avs.scitation.org/journal/jva

J. Vac. Sci. Technol. A 38(2) Mar/Apr 2020; doi: 10.1116/1.5135911 38, 022603-3

Published under license by AVS.

compares the fluorination of Al2O3 with NF3 gas and NF3 induc-tively coupled plasma. The experiments were conducted in a stain-less steel reactor. The pressure was 100 mTorr and the sourcepower was 400W in the case of plasma fluorination. For thermalfluorination, the surface fluorine concentration was 15 at. % andshowed no detectable temperature dependence. As the attenuationlength for photoelectrons stemming from fluorine atoms is fourtimes larger than the penetration depth of fluorine into Al2O3,

17

the fluorine signal can be used as a good approximation for theactual fluorine concentration.

Plasma fluorination using NF3 achieved fluorine concentrationsof up to 41 at. % at 120 °C. The concentration dropped to 20% at265 °C approaching the value of thermal fluorination. Unlike in athermal process, plasma can deliver fluorine in ion and radical formto the substrate surface thereby greatly increasing the reactivity withthe surface. As a result, a far greater fluorine amount is adsorbedinitially than in a purely thermal process. Higher substrate tempera-ture, however, increases the surface desorption rate of physicallybonded fluorine in agreement with the measured concentration dropreported above. This temperature behavior of plasma fluorination isdifferent from the previously reported thermal fluorination wherediffusion of fluorine into the subsurface lattice was found to acceler-ate with the temperature.13 While plasma appears promising for

deeper fluorination and to achieve larger EPC, we find that usingplasma in the fluorination step severely compromised selectivity tomask materials such as Si3N4 or SiO2.

Thermal fluorination with NF3 was not effective in the large-scale reactor featuring alumina and yttria walls. The fluorinesurface concentrations were lower and exhibited a large sensitivityto various wall treatments with plasma and were not repeatable asindicated by the large error bar in Fig. 7. The fluorine concentrationwas, on average, below 10% resulting in negligible EPCs whenDMAC was applied. We attribute the difference in thermal fluorina-tion with NF3 to the difference in reactor wall materials (stainlesssteel versus Y2O3). Sensitivity of the ALE process with respect to thesurrounding surfaces was previously reported for Al2O3 ALE withHF/TMA.18 Hennessy et al. did not partition whether the sensitivityto LiF near the wafer was caused by the HF or the TMA step.18

Nevertheless, there is emerging evidence that reactor wall materialsmust be taken into consideration when comparing EPC’s of thesame thermal ALE chemistry.

Fluorination with HF yielded fluorine surface concentrationswell above the etching threshold, and the results were repeatable.A close analysis of the Al 2p XPS peak at 75.5 eV did confirm the

FIG. 6. Fluorine concentration at the surface of Al2O3 after fluorination with NF3gas and plasma as a function of temperature. FIG. 7. Fluorine concentration at the surface of Al2O3 after fluorination with NF3

and anhydrous HF gas for two different reactors. A: Small scale reactor withstainless steel walls; B: 300 mm reactor with Y2O3 covered walls.

FIG. 8. Saturation curves for thermalALE of Al2O3 with HF/DMAC at30 mTorr and 250 °C with DMAC steptime kept constant at 30 s (a) and HFstep time kept constant at 30 s (b).

ARTICLE avs.scitation.org/journal/jva

J. Vac. Sci. Technol. A 38(2) Mar/Apr 2020; doi: 10.1116/1.5135911 38, 022603-4

Published under license by AVS.

presence of aluminum oxyfluoride but displayed no discernableevidence that the formation of AlF3 had taken place. This resultis in agreement with Cano et al.13 Both XPS measurements wereconducted with air exposure between fluorination and analysis.Therefore, we cannot completely exclude the existence of incipientAlF3, which is converted back to an oxyfluoride during air exposure(prior to XPS).

C. Thermal ALE of Al2O3

Thermal ALE of Al2O3 is achieved by combining the HFfluorination step with DMAC removal of AlF3. The condition ofapparent self-limitation of both steps is met after about 30 s of gasflow (see Fig. 8) at a temperature of 250 °C with the entire processbeing run at a constant pressure of 30mTorr. The resulting EPC was0.14 nm. Previously, EPC values of 0.032 nm have been reported forALD deposited Al2O3 at the same temperature.10 Differencesbetween the two experiments include the Al2O3 film preparation, thefluorination precursor (HF-pyridine versus anhydrous HF), pressure(partial versus absolute pressure) and reactor wall materials(alumina-coated stainless steel versus yttria-coated aluminum).

Figure 9 shows the evolution of the fluorine surface concentra-tion during exposure to DMAC. The fluorine surface concentration

dropped from 14% to 9% after 90 s treatment with DMAC at30 mTorr. The control value of a sample, which was not treatedwith HF, was 6%. At the same time, the oxygen concentrationincreased from 50% to 54%. These findings are consistent with agradual re-establishing of the original Al2O3 surface under DMACexposure by removing the fluorinated top layer.

As AlF3 was consumed entirely by thermal etching withDMAC (see Fig. 2), we conclude that the prevalent material on thesurface after thermal fluorination must be an aluminum oxyfluor-ide and cannot be AlF3. It is possible that residual fluorine isbonded to aluminum, which is also bonded to oxygen, that fluorineis not bonded to aluminum at all (F2 or HF), or that it is bonded tofully fluorinated aluminum, which is surrounded by aluminumoxide compounds and not accessible to DMAC.

Figure 10 shows the fluorine surface concentration immedi-ately after HF treatment as a function of temperature. We noticedthat fluorine concentration with HF was generally higher comparedto the thermal values achieved with NF3 likely due to the higherreactivity of HF. However, like in the plasma case described aboveand in Fig. 6, the concentrations dropped with increasing tempera-ture (unlike in the thermal NF3 case), approaching 15%, again dueto the increased surface desorption rates at a higher temperature.

FIG. 9. Fluorine (a) and oxygen (b)concentrations at an Al2O3 surfaceafter HF exposure and subsequentDMAC exposure as a function ofDMAC exposure time.

FIG. 10. Fluorine concentration after HF exposure and EPC of Al2O3 ALE withHF/DMAC as a function of wafer temperature.

FIG. 11. EPC for HF/DMAC ALE for Al2O3, HfO2, SiO2, and amorphous siliconat 250 °C as function of pressure.

ARTICLE avs.scitation.org/journal/jva

J. Vac. Sci. Technol. A 38(2) Mar/Apr 2020; doi: 10.1116/1.5135911 38, 022603-5

Published under license by AVS.

A nonzero EPC was observed above 170 °C, which is very similarto the etching threshold observed for thermal etching of AlF3 withDMAC. Hence, the lower limit of the ALE window is determinedby the reactivity of DMAC. EPC increases strongly with the tem-perature despite a decreasing fluorine surface concentration afterthe HF step. We attribute the increase in EPC with temperature toan increasing reactivity of DMAC with aluminum oxyfluoride. Thisresponse to temperature should be exponential as with pure AlF3shown in Fig. 4 and should compensate for the lowered fluorineconcentration at higher temperatures.

Thermal ALE with HF/DMAC etched Al2O3 and HfO2 withsimilar EPCs but no etching of SiO2 nor amorphous silicon wasfound for pressures between 10 and 110 mTorr as shown in Fig. 11.The pressure was the same in both steps. DuMont et al. reportedetching of SiO2 with HF/TMA at 300 °C via a conversion reactionfor pressures above 100 mTorr and particularly above 1 Torr.6

We did not observe this mechanism for HF/DMAC for pressures ofup to 110 mTorr. The EPCs for Al2O3 and HfO2 increased by over50% from 10 to 110 mTorr. We cannot discern from our datawhether the HF or the DMAC step caused this increase because thepressure in both steps was increased. Cano et al. reported anincrease in fluoride thickness with pressure with a steep increasebelow 1 Torr and a flattening curve above 4 Torr.13

Etch selectivities do not always transfer from tests withindividual coupons to wafers where all materials are colocated.This is a known etching effect called microloading.19 Based on thesensitivity of the process to the reactor walls, we must assume thatit also exists in thermal isotropic ALE. To test this, 26 nm Al2O3

was deposited on SiO2 and patterned with 45 nm Si3N4. The holeopenings were 60 nm. The samples were etched in a viscous flowreactor at the University of Colorado20 for 500 cycles at 300 °Cat 25 mTorr partial pressure. The etch uniformity from the frontto the back of the reactor is estimated to be better than 10%(min-max) based on ALD results. We could not detect any Si3N4

or SiO2 loss in the cross-sectional electron micrograph shown inFig. 12. The isotropic nature of the process is clearly visible fromthe lateral etching. A more detailed analysis of the vertical andnormal etching amount will be the subject of future studies.

IV. CONCLUSIONS

DMAC gas etches AlF3 in a continuous process at tempera-tures above 180 °C with an activation energy of 1.2 eV. We usedthis reaction to realize a thermal ALE process where Al2O3 isconverted to aluminum oxyfluoride and potentially AlF3 via

exposure to fluorine containing gases and subsequently removedby DMAC. We found fluorination with HF more repeatable andless sensitive to chamber wall materials compared to NF3. Etchingwas detected for the HF/DMAC process for temperatures above170 °C and increases with temperature despite lower fluorine concen-tration. We attribute this behavior to an increased reactivity ofDMAC with aluminum oxyfluoride.

Very high selectivity to SiO2 and Si3N4 was measured both onblanket and patterned wafers. HfO2, however, etched with a similarEPC as Al2O3. Re-entrant profiles demonstrate the isotropic natureof this thermal ALE process.

Over 40 years after John Coburn pioneered RIE, new and usefuletching technologies are still being discovered. Thermal isotropicALE leverages the body of knowledge of the etching and ALD com-munities. John Coburn’s emphasis on etching fundamentals is aninspiration for new generations of researchers and practitioners ofthis essential integrated circuit processing technology.

ACKNOWLEDGMENTS

The authors would like to thank Emir Gurer and Zinki Gargfor the XPS measurements and analysis of the data, Paul Lemairefor preparing the AlF3 films, and Richard Janek for the maskopen processing.

REFERENCES1J. W. Coburn, Appl. Phys. A 59, 451 (1994).2J. W. Coburn and H. F. Winters, J. Appl. Phys. 50, 3189 (1979).3H. F. Winters, J. Appl. Phys. 49, 5165 (1978).4K. J. Kanarik, T. Lill, E. A. Hudson, S. Sriraman, S. Tan, J. Marks, V. Vahedi,and R. A. Gottscho, J. Vac. Sci. Technol. A 33, 020802 (2015).5Y. Lee and S. M. George, ACS Nano 9, 2061 (2015).6J. W. DuMont, A. E. Marquardt, A. M. Cano, and S. M. George, ACS Appl.Mater. Interfaces 9, 10296 (2017).7Y. Lee and S. M. George, Chem. Mater. 29, 8202 (2017).8J. K. C. Chen, N. D. Altieri, T. Kim, E. Chen, T. Lill, M. Shen, and J. P. Chang,J. Vac. Sci. Technol. A 35, 05C305 (2017).9E. Mohimi, X. I. Chu, B. B. Trinh, S. Babar, G. S. Girolami, and J. R. Abelson,ECS J. Solid State Sci. Technol. 7, 491 (2018).10Y. Lee, C. Huffman, and S. M. George, Chem. Mater. 28, 7657 (2016).11Y. Lee, J. W. DuMont, and S. M. George, Chem. Mater. 27, 3648 (2015).12Y. Lee, J. W. DuMont, and S. M. George, J. Phys. Chem. C 119, 25385(2015).13A. M. Cano, A. E. Marquardt, J. W. DuMont, and S. M. George, J. Phys.Chem. C 123, 10346 (2019).14B. E. Deal and A. S. Grove, J. Appl. Phys. 36, 3770 (1965).

FIG. 12. Top-down SEM of a samplebefore Si3N4 mask open and resiststrip (a) and cross section of the samesample after 1000 ALE cycles (b).

ARTICLE avs.scitation.org/journal/jva

J. Vac. Sci. Technol. A 38(2) Mar/Apr 2020; doi: 10.1116/1.5135911 38, 022603-6

Published under license by AVS.

15J. B. Carter, J. P. Holland, E. Peltzer, B. Richardson, E. Bogle, H. T. Nguyen,Y. Melaku, D. Gates, and M. Ben-Dor, J. Vac. Sci. Technol. A 11, 1301 (1993).16Y. Lee and S. M. George, J. Phys. Chem. C 123, 18455 (2019).17A. Fischer, R. Janek, J. Boniface, T. Lill, K. J. Kanarik, Y. Pan, V. Vahedi, andR. A. Gottscho, Proc. SPIE 10149, 101490H (2017).

18J. Hennessy, C. S. Moore, K. Balasubramanian, A. D. Jewell, K. France, andS. Nikzad, J. Vac. Sci. Technol. A 35, 041512 (2017).19C. Hedlund, H.-O. Blom, and S. Berg, J. Vac. Sci. Technol. A 12, 1962 (1994).20J. W. Elam, M. D. Groner, and S. M. George, Rev. Sci. Instrum. 73, 2981(2002).

ARTICLE avs.scitation.org/journal/jva

J. Vac. Sci. Technol. A 38(2) Mar/Apr 2020; doi: 10.1116/1.5135911 38, 022603-7

Published under license by AVS.